透過您的圖書館登入
IP:3.238.6.55
  • 學位論文

發展颱洪時期即時洪水及淹水預報模式

Development of real-time forecasting models for flood and inundation during typhoon periods

指導教授 : 林國峰

摘要


颱風期間劇烈之降雨常造成洪水及淹水而導致嚴重之損害,為了減少洪水及淹水災害,發展預報模式輔助預警系統為常用之方法。支援向量機(Support vector machine, SVM)模擬水文過程(hydrological process)的潛力已經被許多研究所肯定,然而,以支援向量機為基礎之模式在使用上,對於資料的品質受到相當之影響,特別是模式對於極端洪水事件之預報往往無法產生合理之結果。此外傳統支援向量機為基礎之模式為集總模型(lumped model),僅能輸出點預報(或推估),而無法輸出區域預報(或推估)。因此,本論文之目的為建構新型之預報模式來改良傳統模式之缺點,以增進模式之表現。本論文內容依據所提出模式用於水文系統之洪水與淹水兩個問題而分為兩個部份,分述如下: 於論文的第一個部份,本研究發展一改良式洪水預報模式來改善時洪水預報準確性。提出之模式包含了兩個單元:以自組織映射圖(Self-organizing map, SOM)為基礎之群集分析方法(SOM based clustering method)與以SVM為基礎之計算方法(a SVM based computational method)進行多變量非線性迴歸工作。首先,使用SOM群集分析方法去萃取對洪水預報有用之資訊,藉由重製過程將擷取之資訊與觀測資料用於建構洪水預報模式。將本研究提出之預報模式實際應用來呈顯模式之優點,並將提出之模式預報結果與傳統SVM模式作比較,結果顯示所提出之模式表現優於傳統SVM模式。特別針對極端洪水事件,結果顯示所提出之模式可以針對洪水尖峰降低預報誤差,而模式採用SOM為基礎之群集分析結合SVM計算方法所得之預報結果有較佳之表現。 於論文的第二個部份,本研究發展一即時區域淹水預報模式產生未來1至3小時之淹水地圖。提出之模式包含三個單元;分類、點預報、空間擴展。首先,使用K-means群集分析方法,找出淹水資料間之相互關係,並識別出各類別之控制點。接著對於各類別之控制點分別以SVM建構點預報單元產生預報淹水深度。最後依據點預報單元之預報淹水深度及地理資訊以SVM建構空間擴展單元,將點預報擴展至空間預報。將本研究提出之預報模式實際應用來呈顯模式之優點,結果顯示所提出之點預報單元可以準確的預報控制點之淹水深度,且提出之模式可以準確的產生未來1至3小時之淹水地圖。準確的長延時預報可以有效的提升防災預警系統之應變時間,且提出模式為一具效率性之計算方法較適合整合即時資料用以輔助決策系統,而本研究提出之模式對於淹水預警系統有相當之助益。

並列摘要


During typhoon and storm events, floods and inundations caused by torrential rain frequently lead to serious disasters. To mitigate disasters due to floods and inundations, developing forecasting model to support early warning systems is a commonly used measure. Previous researches have shown the potential of the support vector machine (SVM) for modeling hydrological process. However, because SVM based models are nonlinear data-driven approaches, both the quantity and the quality of data available have great influence on the model performance. SVM based models are usually unable to yield satisfactory solutions of extreme values in the flood. In addition, the conventional SVM based models are lumped models which could only produce point forecasts (or estimates) rather than regional forecasts (or estimates). In this thesis, novel approach to the construction of the forecasting model for solving the conventional problems is proposed. Two studies are conducted herein to demonstrate the superiority of the proposed model. In the first study, an improved flood forecasting model is proposed to obtain more accurate hourly flood forecasts. The proposed model contains two parts: a self organizing map (SOM) based clustering method and a SVM based computational method. Firstly, the SOM is used to analyse observed data to extract data with specific properties, which are capable of providing valuable information for flood forecasting. After reprocessing, these extracted data and the observed data are used to construct the SVM based model. An actual application is conducted to clearly demonstrate the advantage of the proposed model. The comparison between the proposed model and the conventional SVM model, which is constructed without SOM, is performed. The results indicate that the proposed model is better performed than the conventional SVM model. Moreover, as regards the extreme events, the result shows that the proposed model reduces the forecasting error, especially the error of peak flood. It is confirmed that because of the use of data extracted by SOM, the improved forecasting performance is obtained. In the second study, a real-time regional forecasting model is proposed to yield 1- to 3-h lead time inundation maps. The proposed model is composed of three parts: classification, point forecasting and spatial expansion. First, the K-means based cluster analysis is developed to group the inundation depths and to indentify the control points. Second, the SVM is used as the computational method to develop the point forecasting module to yield inundation forecasts for each control point. Third, based on the forecasted depths at the control points and the geographic information of the forecast grids, the spatial expansion module is developed to expand the point forecasts to the spatial forecasts. An actual application is conducted to clearly demonstrate the advantage of the proposed model. The results indicate that the point forecasting module can generate accurate point inundation forecasts for each control point, and the proposed model can provide accurate inundation maps for 1- to 3-h lead times. The accurate long lead time forecasts can extend the lead time for issuing warnings to allow sufficient time for taking emergency measures. Furthermore, the proposed model is an efficient process that can be trained rapidly with real-time data and is more suitable to be integrated with the decision support system. In conclusion, the proposed modeling technique is expected to be useful to support the inundation warning systems.

參考文獻


Adamowski J, Chan HF, Prasher SO, Sharda, VN. 2012. Comparison of multivariate adaptive regression splines with coupled wavelet transform artificial neural networks for runoff forecasting in Himalayan micro-watersheds with limited data. Journal of Hydroinformatics 14(3): 731–744.
ASCE Task Committee on Application of Artificial Neural Networks in Hydrology. 2000b. Artificial Neural Networks in hydrology, II: Hydrological Applications. Journal of Hydrologic Engineering, ASCE 5(2): 124–137.
Babovic V. 2005. Data mining in hydrology - Invited commentary. Hydrological Processes 19(7): 1511–1515.
Babovic V, Keijzer M. 2006. Rainfall‐Runoff Modeling Based on Genetic Programming. Encyclopedia of Hydrological Sciences. DOI: 10.1002/0470848944.hsa017.
Bates PD, De Roo APJ. 2000. A simple raster-based model for flood inundation simulation. Journal of Hydrology 236(1–2): 54–77.

延伸閱讀